PD: glb_top silently failing LVS?

[steveri]

It looks as though glb_top has been failing lvs for quite some time now, maybe months. See e.g. glb_top log for build 365 (
https://buildkite.com/tapeout-aha/fullchip/builds/365 )

E   FileNotFoundError: [Errno 2] No such file or directory: 'lvs.report'
=========================== short test summary info ============================
FAILED mflowgen-check-postconditions.py::test_0_ - FileNotFoundError: [Errno ...
1 failed in 0.08s
 
make[1]: *** [23-mentor-calibre-lvs/.postconditions.stamp] Error 1
make[1]: Leaving directory `/build/gold.365/full_chip/16-glb_top'

Unfortunately, although the child make-script ("make lvs") fails, the parent make script ("make glb_top") does not propagate the error to its level, so it never got caught by my CI script. Also, it's not a "real" failure, it's just that LVS is missing one of the requisite files needed to do a proper comparison, so it complains and dies. The later whole-chip LVS has access to all necessary files and, since glb_top apparently does not have any real discrepancies, the top-level LVS passes.

The problem appears to be an incorrectly named input for the glb-top LVS step, a matter that is addressed by mflowgen issue 122 ( mflowgen/mflowgen#122 ) and an upcoming garnet pull. Briefly, the lvs step asks for sram.spi whereas the actual filename at this point is glb_tile_sram.spi, resulting in this complaint from the LVS process:

// Calibre Interactive Error: Source netlist references but does not define 1 subckt: TS1N16FFCLLSBLVTC2048X64M8SW

This issue is mainly just an FYI and a request-for-comments, since (I think) the problem is being dealt with. Your input is welcome! Once everything is fully vetted, I will close this issue again.